Distributed intelligent power distribution network self-healing control method and system

Abstract

The present application relates to power distribution system technical field, especially to distributed intelligent power distribution network self-healing control method and system.The method comprises the following steps: step S1, real-time collection of distributed intelligent power distribution network operation data; step S2, fault detection; step S3, parallel protection control; step S4, building fault section positioning model and island division model, and positioning fault section according to fault section positioning model, and dividing island according to island division model; step S5, building double-target optimization model; step S6, solving double-target optimization model according to improved ant colony algorithm, and obtaining optimal switch action table; step S7, safety check of optimal switch action table, and self-healing control of distributed intelligent power distribution network according to safety check result; step S8, feedback of self-healing control situation of distributed intelligent power distribution network.The present application improves fault recovery speed, accuracy and reliability of distributed intelligent power distribution network.

Description

Technical Field

[0001] This invention relates to the field of power distribution system technology, and in particular to a self-healing control method and system for distributed smart power distribution networks. Background Technology

[0002] With the ongoing energy transition, the penetration rate of distributed generation in distribution networks is constantly increasing, and load types are becoming increasingly diverse. This has transformed distribution networks from traditional passive radial networks into complex active networks with multiple power sources and bidirectional power flow. This fundamental change has brought severe challenges to the operation and control of distribution networks, especially their fault handling capabilities. Traditional distribution network fault handling mainly relies on centralized processing and local protection modes. Existing methods, when restoring power supply, often only aim to maximize the restored power supply range, lacking comprehensive optimization of the voltage quality and economic operation of the grid after restoration. At the same time, traditional optimization algorithms suffer from slow convergence speed and susceptibility to local optima when dealing with large-scale, multi-constraint distribution network reconfiguration problems, failing to meet the real-time requirements of second-level self-healing. Therefore, there is an urgent need in this field for a distributed intelligent self-healing control scheme that can adapt to scenarios with a high proportion of distributed generation access, possesses rapid response, intelligent decision-making, and collaborative control capabilities, and can ensure high-quality and economical grid operation while restoring power supply.

[0003] Chinese Patent Publication No. CN119994892A discloses a self-healing control method, system, and terminal for distribution network faults, relating to the field of automatic control technology for distribution networks. The key technical points are: the self-healing control strategy in this invention only includes the distribution area where power is restored and the fault location where power continues to be cut off, improving the efficiency of fault analysis; furthermore, the distribution terminal analyzes the correlation between the distribution area and the fault location based on the received self-healing control strategy, thereby determining whether a safety check should be performed when power is restored to the distribution area. However, this solution relies heavily on centralized decision-making and remote control by the distribution network master station. Its overall performance is limited by the processing power of the master station and the reliability of the communication network between the master station and the terminal. Moreover, its preset protection and self-healing strategies are difficult to adapt flexibly to complex situations such as variable distribution network operation modes and high distributed power penetration, failing to achieve dynamically optimized control effects, resulting in slow fault recovery speed, low accuracy, and low reliability. Summary of the Invention

[0004] To address this, the present invention provides a self-healing control method and system for distributed smart distribution networks, which overcomes the problems of slow fault recovery speed, low accuracy, and low reliability in existing distributed smart distribution networks.

[0005] To achieve the above objectives, the present invention provides a self-healing control method for distributed smart distribution networks, the method comprising:

[0006] Step S1: Real-time collection of distributed smart distribution network operation data to obtain distributed smart distribution network operation data;

[0007] Step S2: Perform fault detection on the distributed smart distribution network operation data to obtain fault signals;

[0008] Step S3: Perform parallel protection control on the distributed smart distribution network based on the fault signal;

[0009] Step S4: Construct a fault segment localization model and an island partitioning model, locate fault segments according to the fault segment localization model, partition islands according to the island partitioning model, and obtain a fault edge dataset and an island cluster dataset.

[0010] Step S5: Construct a dual-objective optimization model;

[0011] Step S6: Solve the bi-objective optimization model using the improved ant colony algorithm to obtain the optimal switching action table;

[0012] Step S7: Perform a safety check on the optimal switch action table, and perform self-healing control on the distributed smart distribution network based on the safety check results;

[0013] Step S8: Feedback is provided on the self-healing control status of the distributed smart distribution network.

[0014] Further, in step S2, when performing fault detection on the distributed smart distribution network operation data, the distributed smart distribution network operation data is input into the fault detection model. The fault detection model outputs current surge I2 and voltage sag value U2, and compares the current surge I2 and voltage sag value U2 with preset current surge I0 and preset voltage sag value U0, respectively. Based on the comparison results, the fault occurrence is judged, and a fault signal is output based on the judgment result, wherein:

[0015] When the comparison result is I2≤I0 and U2≥U0, the fault occurrence is determined to be no fault, the subsequent operation is stopped, and steps S1-S2 are repeated until the fault occurrence is a fault, at which point the distributed smart distribution network operation data corresponding to the fault occurrence is output as a fault signal.

[0016] When the comparison result is not I2≤I0 and U2≥U0, the fault situation is determined to be a fault, and the distributed smart distribution network operation data corresponding to the fault situation is output as a fault signal.

[0017] Further, in step S3, when performing parallel protection control of the distributed smart distribution network based on the fault signal, the fault signal is input into the fault current identification model for fault current value identification to obtain the fault current value Ig. The fault current value Ig is compared with the preset fault current value Ig0. Based on the comparison result, the emergency situation of the fault is judged, and the parallel protection control strategy is output based on the judgment result, wherein:

[0018] When Ig≥Ig0, the fault emergency situation is determined to be emergency, and the output parallel protection control strategy is to issue a trip command;

[0019] When Ig < Ig0, the emergency situation of the fault is determined to be non-emergency, and the output parallel protection control strategy is to issue a deep reinforcement learning control command.

[0020] In step S3, the deep reinforcement learning control instructions include: inputting the fault signal into the fault instantaneous trip model, outputting the instantaneous trip setting value of the line protection from the fault instantaneous trip model, and transmitting the instantaneous trip setting value of the line protection to the digital protection relay.

[0021] Further, in step S4, a fault segment location model is constructed. When locating a fault segment based on the fault segment location model, the historical fault segment location database is divided into an 80% segment training set and a 20% segment test set. The segment training set is input into a graph neural network model to train the graph neural network model, resulting in a trained graph neural network model. The segment test set is input into the trained graph neural network model for testing until the segment testing accuracy Q of the trained graph neural network model reaches 95%. Then, the trained graph neural network model is output as the fault segment location model, and the fault signal is input into the fault segment location model to obtain the fault edge dataset.

[0022] Further, in step S4, an island partitioning model is constructed. When partitioning islands according to the island partitioning model, the island partitioning database is divided into a 70% island training set, a 15% island validation set, and a 15% island test set. The island training set is input into a convolutional neural network model to train the convolutional neural network model, resulting in a trained convolutional neural network model. The island validation set is input into the trained convolutional neural network model for iterative optimization, resulting in a validated convolutional neural network model. The island test set is then input into the validated convolutional neural network model until the island test accuracy G of the validated convolutional neural network model reaches 98%. At this point, the validated convolutional neural network model is output as a fault island partitioning model, and the distributed smart distribution network operation data is input into the fault island partitioning model to obtain a dataset of possible island clusters.

[0023] Furthermore, in step S5, when constructing the dual-objective optimization model, "minimum node voltage deviation after fault" and "minimum network loss" are taken as the optimization objectives of the dual-objective optimization model. "Branch current does not exceed the thermal stability limit", "node voltage is maintained between 0.95pu and 1.05pu", "distributed power output does not exceed its rated capacity", and "the number of switching actions k is limited to k≤K0 times" are taken as the constraints of the dual-objective optimization model. The fault edge dataset and the islandable cluster dataset are taken as the inputs of the dual-objective optimization model to obtain the dual-objective optimization model. Here, the dual-objective optimization model is the mathematical optimization model to be solved, and K0 is the preset limit number of times.

[0024] Further, in step S6, when solving the bi-objective optimization model using the improved ant colony algorithm, the bi-objective optimization model is solved using the ant colony solution method based on the improved ant colony algorithm. The ant colony solution method includes:

[0025] Step Y01, Initialization Phase: Using the number of network nodes N as the independent variable, according to ρ(N) = 0.90 – 0.40·e (–N / 50) Calculate the evaporation coefficient ρ(N) and assign values ​​to the pheromone H0(i,j) of all switch edges;

[0026] Step Y02, Ant Path Construction Phase: Release 32 ants in parallel, according to probability. Select the next switch state to form a candidate reconstruction scheme S K ,in, , allowed is the set of switches that are allowed to operate;

[0027] Step Y03, Local Search Phase: Generating candidate reconstruction schemes S for each ant K If the node voltage increases and the network loss decreases after performing the 2-opt switch operation, then the new solution replaces the original solution.

[0028] Step Y04, Fitness Evaluation Phase: Linearized power flow calculation is used to calculate the voltage over-limit penalty Mv and network loss P. loss And according to the fitness function F=λ·Mv+(1-λ)·P loss Calculate the fitness value F for each candidate solution;

[0029] Step Y05, Pheromone Update Phase: Additional pheromone ΔH is released for the globally optimal path, and ΔH = Q / F. best Where Q is a constant, F best The current optimal fitness value is set, and the pheromone matrix is ​​updated according to the adaptive evaporation coefficient ρ(N,t);

[0030] Step Y06, Termination Judgment Stage: When the improvement rate R of the best solution for 5 consecutive generations is less than 0.1%, output the optimal switching action table S*.

[0031] Furthermore, in step S6, when solving the bi-objective optimization model using the improved ant colony algorithm, environmental data and hardware data are also acquired. The environmental data includes real-time solar radiation ratio (SR), wind speed (WS), and relative humidity (RH). The hardware data includes edge box temperature (T). gpu The environmental and hardware data are input into the evaporation coefficient effective probability evaluation model to obtain the evaporation coefficient effective probability C. The evaporation coefficient effective probability C is compared with the preset evaporation coefficient effective probability C0. Based on the comparison result, the effectiveness of the evaporation coefficient is judged, and the evaporation coefficient ρ(N) is corrected according to the judgment result.

[0032] When C≥C0, the evaporation coefficient is considered valid and no correction is made to the evaporation coefficient ρ(N);

[0033] When C < C0, the effective case of the evaporation coefficient is determined to be invalid, and the evaporation coefficient ρ(N) is corrected. The corrected evaporation coefficient is then set to ρ(N)'. And the ant colony method is executed based on the corrected evaporation coefficient ρ(N)`.

[0034] Further, in step S8, when providing feedback on the self-healing control status of the distributed smart distribution network, the distributed smart distribution network operation data after the corresponding smart circuit breaker executes the operation according to the GOOSE control frame is collected in real time through the micro synchronous phasor measurement unit and the distribution terminal unit. The distributed smart distribution network operation data after the corresponding smart circuit breaker executes the operation according to the GOOSE control frame, along with the fault signal, are input into the self-healing performance index evaluation model to obtain an evaluation score Z. The evaluation score Z is compared with the preset evaluation score Z0. Based on the comparison result, the self-healing performance is judged, and the ant colony solution method is optimized based on the judgment result. Wherein:

[0035] When Z≥Z0, the self-healing performance is deemed satisfactory, and the ant colony solution method is not optimized.

[0036] When Z < Z0, the self-healing performance is deemed substandard. The ant colony algorithm is then optimized. The optimization involves obtaining the number of network nodes N in the ant colony algorithm and comparing it with the preset number of network nodes N0. Based on the comparison result, the network node size is determined, and the corrected evaporation coefficient ρ(N)` is adjusted accordingly.

[0037] When N < N0, the network node size is determined to be small, and the corrected evaporation coefficient ρ(N)` is not adjusted.

[0038] When N≥N0, the network node size is determined to be large-scale. The corrected evaporation coefficient ρ(N)` is adjusted, and the fault signal is input into the fault risk probability assessment model to obtain the fault risk probability score P. fault According to the failure risk probability score P fault The corrected evaporation coefficient ρ(N)` is adjusted to obtain the adjusted evaporation coefficient ρ(N)``, and the ant colony algorithm is performed based on the adjusted evaporation coefficient ρ(N)``, where:

[0039] ;

[0040] It is a step function;

[0041] In P fault When ≥0.8, ;

[0042] In P fault When <0.8, .

[0043] On the other hand, the present invention also provides a distributed smart distribution network self-healing control system, the system comprising:

[0044] The data acquisition module is used to collect real-time operation data of the distributed smart distribution network to obtain the distributed smart distribution network operation data;

[0045] The fault detection module is used to detect faults in the operation data of the distributed smart distribution network and obtain fault signals.

[0046] The parallel control module is used to perform parallel protection control on the distributed smart distribution network based on fault signals.

[0047] The segment island module is used to construct a fault segment location model and an island partitioning model. It performs fault segment location based on the fault segment location model and island partitioning based on the island partitioning model, resulting in a fault edge dataset and a potentially isolated cluster dataset.

[0048] The ant colony optimization module is used to construct a dual-objective optimization model, solve the dual-objective optimization model according to the improved ant colony algorithm to obtain the optimal switching action table, perform safety verification on the optimal switching action table, and perform self-healing control on the distributed smart distribution network based on the safety verification results.

[0049] The self-healing feedback module is used to provide feedback on the self-healing control status of the distributed smart distribution network.

[0050] Compared with existing technologies, the beneficial effects of this invention are as follows: The method, through step S1, deploys a miniature synchronous phasor measurement unit and a distribution terminal unit, thereby providing a unified, accurate, and highly reliable data foundation. The method, through step S2, completes fault judgment based on the dual criteria of "current surge" and "voltage sag," facilitating rapid and reliable initial screening of faults and improving the fault recovery speed of smart distribution networks. Simultaneously, the mutual verification of the dual criteria effectively avoids false tripping caused by load switching, lightning interference, etc., ensuring that subsequent self-healing processes are accurately initiated only under real fault conditions. The method, through step S3, uses deep reinforcement learning to prevent cascading tripping control and the parallel operation of traditional instantaneous overcurrent protection, forming a "double insurance" mechanism, thus combining speed and selectivity to improve the intelligence and reliability of protection. The method, through step S4, uses graph neural networks and convolutional neural networks to learn models, achieving precise location of fault sections and islanding of areas with a high proportion of photovoltaic power, facilitating fault handling from... The method elevates the "isolation" level to the "recovery" planning level, providing crucial input for subsequent recovery optimization. Step S5 transforms fuzzy decision-making driven by engineering experience into a scientific optimization problem with clear objectives and constraints, providing precise input for Step S6 and ensuring the overall optimality of the final recovery solution. Step S6 deeply improves the traditional ant colony algorithm to efficiently solve the dual-objective optimization model in Step S5. Simultaneously, by incorporating environmental and hardware data feedback, it enhances adaptability and robustness in real industrial environments. Step S7 utilizes the millisecond-level transmission characteristics of GOOSE frames to achieve a secure and efficient closed loop from the computational space to the physical world. Step S8 evaluates the actual effect of each self-healing action and feeds back the evaporation coefficient from Step S6, adaptively adjusting strategies for networks of different sizes and fault severities to continuously improve the decision-making quality and robustness of future self-healing actions. Attached Figure Description

[0051] Figure 1 This is a flowchart illustrating the self-healing control method for distributed smart distribution networks in this embodiment.

[0052] Figure 2 This is a schematic diagram of the structure of the self-healing control system for the distributed smart distribution network in this embodiment. Detailed Implementation

[0053] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0054] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0055] It should be noted that in the description of this invention, the terms "upper", "lower", "left", "right", "inner", "outer", etc., which indicate directions or positional relationships, are based on the directions or positional relationships shown in the accompanying drawings. This is only for the convenience of description and is not intended to indicate or imply that the device or element must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation of this invention.

[0056] Furthermore, it should be noted that, in the description of this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0057] Please see Figure 1 As shown, this is a flowchart illustrating the self-healing control method for distributed smart distribution networks in this embodiment. The method includes:

[0058] Step S1: Real-time collection of distributed smart distribution network operation data to obtain distributed smart distribution network operation data;

[0059] Step S2: Perform fault detection on the distributed smart distribution network operation data to obtain fault signals;

[0060] Step S3: Perform parallel protection control on the distributed smart distribution network based on the fault signal;

[0061] Step S4: Construct a fault segment localization model and an island partitioning model, locate fault segments according to the fault segment localization model, partition islands according to the island partitioning model, and obtain a fault edge dataset and an island cluster dataset.

[0062] Step S5: Construct a dual-objective optimization model;

[0063] Step S6: Solve the bi-objective optimization model using the improved ant colony algorithm to obtain the optimal switching action table;

[0064] Step S7: Perform a safety check on the optimal switch action table, and perform self-healing control on the distributed smart distribution network based on the safety check results;

[0065] Step S8: Feedback is provided on the self-healing control status of the distributed smart distribution network.

[0066] Specifically, the method is applied to distributed smart distribution network terminals. It achieves continuous self-optimization and self-healing control of the distributed smart distribution network through real-time data acquisition, fault detection, parallel protection, intelligent analysis, model building, intelligent solution, secure execution, and feedback optimization. In particular, step S1 deploys miniature synchronous phasor measurement units and distribution terminal units, providing a unified, accurate, and highly reliable data foundation. Step S2 uses a dual criterion of "current surge" and "voltage sag" to complete fault judgment, enabling rapid and reliable initial fault screening and improving the fault recovery speed of the smart distribution network. Simultaneously, the mutual verification of the dual criteria effectively avoids false tripping caused by load switching, lightning interference, etc., ensuring that subsequent self-healing processes are accurately initiated only under real fault conditions. Step S3 uses deep reinforcement learning to prevent cascading tripping control in parallel with traditional instantaneous overcurrent protection, forming a "double insurance" mechanism that combines speed and selectivity, thereby improving the intelligence and reliability of protection. Step S4 uses graph neural networks and convolutional neural networks to learn... The method employs a model to accurately locate faulty sections and divide areas with high photovoltaic (PV) coverage into isolated islands. This elevates fault handling from the "isolation" level to the "recovery" planning level, providing crucial input for subsequent optimization and recovery. Step S5 transforms fuzzy decision-making driven by engineering experience into a scientific optimization problem with clear objectives and constraints, providing precise input for step S6 and ensuring the overall optimality of the final recovery solution. Step S6 deeply improves the traditional ant colony algorithm to efficiently solve the dual-objective optimization model in step S5. Furthermore, by incorporating environmental and hardware data feedback, it enhances adaptability and robustness in real industrial environments. Step S7 utilizes the millisecond-level transmission characteristics of GOOSE frames to achieve a secure and efficient closed loop from the computational space to the physical world. Step S8 evaluates the actual effect of each self-healing action and feeds back the evaporation coefficient from step S6, adaptively adjusting strategies for networks of different sizes and fault severities to continuously improve the decision-making quality and robustness of future self-healing actions.

[0067] Specifically, in step S1, when the distributed smart distribution network operation data is collected in real time, the micro synchronous phasor measurement units and distribution terminal units deployed at each key node of the distributed smart distribution network collect the distributed smart distribution network operation data in real time to obtain the distributed smart distribution network operation data.

[0068] Specifically, the key nodes of the distributed smart distribution network refer to the physical locations in the distribution network topology that play a decisive role in power flow distribution, fault diagnosis, and power restoration. The micro-synchronous phasor measurement unit refers to a high-precision measurement device that achieves microsecond-level time synchronization across the entire network based on the BeiDou / GPS and IEEE 1588 precision clock protocol. The distribution terminal unit refers to intelligent electronic devices installed in the distribution network switching stations, distribution rooms, and other field locations, such as feeder terminals and distribution terminals. The distributed smart distribution network operation data refers to the collection of electrical and state quantities that characterize the real-time operating status of the power grid, such as three-phase current, three-phase voltage, power, and phase angle.

[0069] Specifically, step S1 provides a unified, accurate, and highly reliable data foundation by deploying a miniature synchronous phasor measurement unit and a power distribution terminal unit.

[0070] Specifically, in step S2, when performing fault detection on the distributed smart distribution network operation data, the distributed smart distribution network operation data is input into the fault detection model. The fault detection model outputs a current surge I2 and a voltage sag U2, and compares the current surge I2 and the voltage sag U2 with preset current surge I0 and preset voltage sag U0, respectively. Based on the comparison results, the fault occurrence is determined, and a fault signal is output based on the determination results, wherein:

[0071] When the comparison result is I2≤I0 and U2≥U0, the fault occurrence is determined to be no fault, the subsequent operation is stopped, and steps S1-S2 are repeated until the fault occurrence is a fault, at which point the distributed smart distribution network operation data corresponding to the fault occurrence is output as a fault signal.

[0072] When the comparison result is not I2≤I0 and U2≥U0, the fault situation is determined to be a fault, and the distributed smart distribution network operation data corresponding to the fault situation is output as a fault signal.

[0073] Specifically, the fault detection model refers to a pre-trained deep convolutional neural network model that meets preset fault detection requirements; the current surge I2 refers to the change in the instantaneous current value between the current sampling point and historical sampling points; the voltage dip value U2 refers to the effective voltage value; and the preset current surge I0 refers to a preset current change value, taken as the rated operating current of the distribution network I... n A certain proportion, for example, I0 = 0.2I n The preset voltage dip value U0 refers to a pre-set voltage drop threshold, expressed as a percentage of the rated voltage Un, for example, U0 = 0.85U. n .

[0074] Specifically, step S2 completes the fault judgment based on the dual criteria of "current surge" and "voltage sag" to facilitate rapid and reliable initial screening of faults, thereby improving the fault recovery speed of the smart distribution network. At the same time, through the mutual verification of the dual criteria, false activation caused by load switching, lightning interference, etc. is effectively avoided, ensuring that the subsequent self-healing process is accurately activated only under real fault conditions.

[0075] Specifically, in step S3, when performing parallel protection control of the distributed smart distribution network based on the fault signal, the fault signal is input into the fault current identification model for fault current value identification to obtain the fault current value Ig. The fault current value Ig is compared with the preset fault current value Ig0. Based on the comparison result, the emergency situation of the fault is judged, and the parallel protection control strategy is output based on the judgment result, wherein:

[0076] When Ig≥Ig0, the fault emergency situation is determined to be emergency, and the output parallel protection control strategy is to issue a trip command;

[0077] When Ig < Ig0, the emergency situation of the fault is determined to be non-emergency, and the output parallel protection control strategy is to issue a deep reinforcement learning control command.

[0078] In step S3, the deep reinforcement learning control instructions include: inputting the fault signal into the fault instantaneous trip model, outputting the instantaneous trip setting value of the line protection from the fault instantaneous trip model, and transmitting the instantaneous trip setting value of the line protection to the digital protection relay.

[0079] Specifically, the fault current identification model refers to a pre-trained machine learning model that meets preset fault current identification requirements. The fault current value Ig refers to the effective value of the fundamental fault current currently flowing through the protection installation point, calculated and output by the fault current identification model. The preset fault current value Ig0 refers to a fixed current value set for the instantaneous overcurrent protection section. The trip command refers to the opening command generated when Ig≥Ig0 is satisfied. The fault instantaneous overcurrent model refers to a pre-trained deep reinforcement learning agent that meets the instantaneous overcurrent setting requirements of the output line protection. The instantaneous overcurrent setting of the line protection refers to the new instantaneous overcurrent setting value output by the fault instantaneous overcurrent model after dynamic adjustment. The digital protection relay refers to a modern protection device with communication and remote setting modification functions, such as the NARI PCS-9617. This embodiment does not limit the transmission method of transmitting the instantaneous overcurrent setting value of the line protection to the digital protection relay. Those skilled in the art can freely set it according to the actual situation, as long as the requirement of transmitting the instantaneous overcurrent setting value of the line protection to the digital protection relay is met.

[0080] Specifically, step S3 establishes a "double insurance" mechanism by combining deep reinforcement learning-based anti-overlapping trip control with traditional instantaneous overcurrent protection, thereby achieving both speed and selectivity to improve the intelligence and reliability of the protection.

[0081] Specifically, in step S4, when constructing a fault segment location model and locating fault segments based on the fault segment location model, the historical fault segment location database is divided into an 80% segment training set and a 20% segment test set. The segment training set is input into a graph neural network model to train the graph neural network model, resulting in a trained graph neural network model. The segment test set is input into the trained graph neural network model for testing until the segment test accuracy Q of the trained graph neural network model reaches 95%. Then, the trained graph neural network model is output as the fault segment location model, and the fault signal is input into the fault segment location model to obtain the fault edge dataset.

[0082] In step S4, an island partitioning model is constructed. When partitioning islands according to the island partitioning model, the island partitioning database is divided into a 70% island training set, a 15% island validation set, and a 15% island test set. The island training set is input into a convolutional neural network model to train the convolutional neural network model, resulting in a trained convolutional neural network model. The island validation set is input into the trained convolutional neural network model for iterative optimization, resulting in a validated convolutional neural network model. The island test set is then input into the validated convolutional neural network model until the island test accuracy G of the validated convolutional neural network model reaches 98%. At this point, the validated convolutional neural network model is output as the fault island partitioning model, and the distributed smart distribution network operation data is input into the fault island partitioning model to obtain a dataset of possible island clusters.

[0083] Specifically, the historical fault segment location database refers to a dataset containing historical and simulated fault cases, including multi-source information at the time of the fault and its corresponding verified actual fault segment labels. The segment training set refers to a dataset used for learning the parameters of the graph neural network model. The segment test set refers to a dataset used to independently evaluate the generalization performance of the graph neural network model. The graph neural network model refers to a deep learning model used to process graph-structured data. The segment test accuracy Q refers to the percentage of fault segments correctly predicted by the graph neural network model on the segment test set out of the total number of test cases. The fault edge dataset refers to the set of line segments identified as faults output by the fault segment location model after inference. The island partitioning database refers to a dataset containing various... The data set of the distribution network status under the operating scenario includes real-time generation, load, energy storage SOC, network topology, and the corresponding optimal islanding scheme labels. The islanding training set refers to the dataset used for learning the parameters of the convolutional neural network model. The islanding validation set refers to the dataset used to select the best convolutional neural network model. The islanding test set refers to the dataset used for the final unbiased evaluation of the convolutional neural network model. The convolutional neural network model refers to a deep learning model that extracts features from data with a grid-like topology. The islanding test accuracy G refers to the percentage of cases where the output islanding scheme on the islanding test set is consistent with the optimal scheme in terms of key indicators. The islandable cluster dataset refers to the set of independently operable power supply areas output by the faulty islanding model after inference.

[0084] Specifically, step S4 uses graph neural networks and convolutional neural networks to learn models, thereby achieving accurate location of faulty sections and islanding of areas with a high proportion of photovoltaics. This facilitates the upgrading of fault handling from the "isolation" level to the "recovery" planning level, providing key input for subsequent optimization and recovery.

[0085] Specifically, in step S5, when constructing the dual-objective optimization model, "minimum node voltage deviation after fault" and "minimum network loss" are taken as the optimization objectives of the dual-objective optimization model. "Branch current does not exceed the thermal stability limit", "node voltage is maintained between 0.95pu and 1.05pu", "distributed power output does not exceed its rated capacity", and "the number of switching actions k is limited to k≤K0 times" are taken as the constraints of the dual-objective optimization model. The fault edge dataset and the islandable cluster dataset are taken as the inputs of the dual-objective optimization model to obtain the dual-objective optimization model. Here, the dual-objective optimization model is the mathematical optimization model to be solved, and K0 is the preset limit number of times.

[0086] Specifically, the constraints refer to the set of all technical restrictions that the solution must satisfy in the optimization problem; the mathematical optimization model to be solved refers to a fully defined mixed integer nonlinear programming problem that includes decision variables, objective functions, and constraints; and the preset limit number of times refers to a positive integer pre-set based on the lifespan and operational risks of the switching machinery, for example, the preset limit number of times K0 = 7 times.

[0087] Specifically, step S5 transforms the fuzzy decision-making driven by engineering experience into a scientific optimization problem with clear objectives and well-defined constraints, providing precise input for step S6 and ensuring the overall optimality of the final recovery solution.

[0088] Specifically, in step S6, when solving the bi-objective optimization model using the improved ant colony algorithm, the bi-objective optimization model is solved using the ant colony solution method based on the improved ant colony algorithm. The ant colony solution method includes:

[0089] Step Y01, Initialization Phase: Using the number of network nodes N as the independent variable, according to ρ(N) = 0.90 – 0.40·e (–N / 50) Calculate the evaporation coefficient ρ(N) and assign values ​​to the pheromone H0(i,j) of all switch edges;

[0090] Step Y02, Ant Path Construction Phase: Release 32 ants in parallel, according to probability. Select the next switch state to form a candidate reconstruction scheme S K ,in, , allowed is the set of switches that are allowed to operate;

[0091] Step Y03, Local Search Phase: Generating candidate reconstruction schemes S for each ant K If the node voltage increases and the network loss decreases after performing the 2-opt switch operation, then the new solution replaces the original solution.

[0092] Step Y04, Fitness Evaluation Phase: Linearized power flow calculation is used to calculate the voltage over-limit penalty Mv and network loss P. loss And according to the fitness function F=λ·Mv+(1-λ)·P loss Calculate the fitness value F for each candidate solution;

[0093] Step Y05, Pheromone Update Phase: Additional pheromone ΔH is released for the globally optimal path, and ΔH = Q / F. best Where Q is a constant, F best The current optimal fitness value is set, and the pheromone matrix is ​​updated according to the adaptive evaporation coefficient ρ(N,t);

[0094] Step Y06, Termination Judgment Stage: When the improvement rate R of the best solution for 5 consecutive generations is less than 0.1%, output the optimal switching action table S*;

[0095] In step S6, when solving the bi-objective optimization model using the improved ant colony algorithm, environmental data and hardware data are also acquired. The environmental data includes real-time solar radiation ratio (SR), wind speed (WS), and relative humidity (RH). The hardware data includes edge box temperature (T). gpu The environmental and hardware data are input into the evaporation coefficient effective probability evaluation model to obtain the evaporation coefficient effective probability C. The evaporation coefficient effective probability C is compared with the preset evaporation coefficient effective probability C0. Based on the comparison result, the effectiveness of the evaporation coefficient is judged, and the evaporation coefficient ρ(N) is corrected according to the judgment result.

[0096] When C≥C0, the evaporation coefficient is considered valid and no correction is made to the evaporation coefficient ρ(N);

[0097] When C < C0, the effective case of the evaporation coefficient is determined to be invalid, and the evaporation coefficient ρ(N) is corrected. The corrected evaporation coefficient is then set to ρ(N)'. And the ant colony method is executed based on the corrected evaporation coefficient ρ(N)`.

[0098] Specifically, the derivation process of the evaporation coefficient ρ(N) in step Y01 is as follows:

[0099] Derivation step Y011, benchmark setting: The theoretical range of the evaporation coefficient ρ is (0,1), and the preset benchmark range is set to [0.50,0.90];

[0100] Derivation step Y012, function form selection: adopt the negative exponential function AB·e (-N / C) Because it can smoothly transition from the initial value to the saturation value, it conforms to the logic of early exploration and later development of the algorithm;

[0101] Derivation step Y013, boundary conditions determined:

[0102] When N→0 (very small scale), e (-0 / C) Since ρ is approximately 1, we want ρ to be small to encourage exploration, so ρmin ≈ AB. Let ρmin ≈ 0.50.

[0103] As N→∞ (ultra-large scale), e -∞ →0, we want ρ to be larger to accelerate convergence, so ρmax≈A, let ρmax≈0.90;

[0104] Derivation step Y014, parameter solution and scaling factor:

[0105] Given A = 0.90 and AB = 0.50, we can solve for B = 0.40.

[0106] C is a scaling factor that controls the rate of change. Based on engineering experience, the scale of a distribution network is often divided by 50 nodes, so we take C=50. That is, when N=50, e -1 ≈0.367, ρ(50)≈0.90-0.40×0.367≈0.753, which is in the middle value, so the final formula is: ρ(N)=0.90–0.40·e (–N / 50) ;

[0107] The probability in step Y02 The derivation process is as follows:

[0108] Derivation step Y021, pheromone term H(i,j) α : Represents historical experience, H(i,j) is the pheromone concentration on edge (i,j), and α is the pheromone weight. The higher the pheromone concentration, the more frequently the edge appears in historical high-quality solutions, and the greater the probability of it being selected.

[0109] Derivation step Y022, heuristic information item η(i,j) β : Represents prior knowledge. In a power distribution network, η(i,j) is often defined as 1 / (Rij·Ploss,ij), where Rij is the line resistance, (Ploss,ij) is the estimated network loss, β is the heuristic information weight, and η(i,j) is the heuristic information weight. β The larger the value, the greater the immediate contribution of choosing that edge to the optimization objective.

[0110] Derivation step Y023, probability normalization: denominator Summing the product of all allowed switches for node i, so that the sum of the probabilities of all outgoing edges is 1, thus forming a probability distribution;

[0111] Derivation step Y024, final formula: ;

[0112] In step Y04, the fitness function is F = λ·Mv + (1-λ)·P loss The derivation process is as follows: A scalarization method using linear weighted sum F = λ·f1 + (1-λ)·f2 is employed, with λ ∈ [0,1]; let f1 represent the voltage over-limit penalty term Mv, and f2 represent the network loss P. loss The voltage over-limit penalty term Mv is used as f1 and network loss P. loss Substituting f2 into the linear weighted sum F = λ·f1 + (1-λ)·f2, we get the final formula: F = λ·Mv + (1-λ)·P loss ;

[0113] The derivation process of the additional pheromone release ΔH in step Y04 of the globally optimal path is as follows:

[0114] Additional pheromone rewards are given to the optimal path found in each generation.

[0115] ΔH should be correlated with the fitness F of the path. In the minimization problem, the smaller the value of F, the higher the quality of the solution. Therefore, ΔH should be inversely proportional to F.

[0116] The simplest inverse relationship is ΔH∝1 / F. A constant Q is introduced as an adjustment coefficient to control the overall magnitude of the additional pheromone release.

[0117] This ΔH applies only to all edges on the globally optimal path, i.e., H(i,j) = H(i,j) + ΔH for all edges (i,j) belonging to the elite path;

[0118] The final formula is: ΔH = Q / F best ;

[0119] In step S6, the derivation process of the corrected evaporation coefficient ρ(N)` is as follows:

[0120] Step S601, Sunlight ratio correction item: ;

[0121] Physical mapping: The real-time solar irradiance ratio SR directly affects photovoltaic output. The lower the SR / SR0, the lower the photovoltaic output is compared to the typical level, which is equivalent to being in a tense state of "energy shortage". SR0 is the preset real-time solar irradiance ratio. When there is an "energy shortage", the convergence speed should be slowed down and more thorough exploration should be carried out to avoid the failure of the scheme due to fluctuations in photovoltaic output. At this time, the evaporation coefficient ρ should be reduced.

[0122] Derivation of the corrected sunshine ratio: When there is insufficient sunlight, the value is positive. The coefficient -0.02 ensures that when the sunlight is 0, that is, when SR=0, the contribution of this term is -0.02, which reduces ρ and enhances the exploration. The 0.02 is determined through simulation experiments.

[0123] Step S602, Wind speed correction item:

[0124] Physical mapping: Wind speed WS affects the heat dissipation capacity of the line. The higher the wind speed, the higher the line current carrying capacity and the more relaxed the network constraints. When the constraints are relaxed, more aggressive convergence can be allowed to find the solution quickly. Therefore, when the wind speed is high, the evaporation coefficient ρ should be increased.

[0125] Derivation of the wind speed correction term:

[0126] With 2 m / s as the reference wind speed, (WS-2) is positive when the wind speed is high. +0.01 means that for every 1 m / s increase in wind speed, ρ increases by 0.01, which moderately accelerates the convergence.

[0127] Step S603, Humidity Correction Item: ;

[0128] Physical mapping: Increased relative humidity (RH) leads to a decrease in the external insulation performance of electrical equipment, increasing the risk of failure. In environments with a high risk of failure, caution should be exercised, and exploration should be strengthened to ensure that a more robust solution is found. Therefore, when the relative humidity (RH) is high, the evaporation coefficient (ρ) should be reduced.

[0129] Humidity correction term derivation: Using 60% as the baseline humidity, (RH-60) / 10 normalizes the change. For example, when the humidity is 70%, the value is 1. The coefficient -0.01 means that for every 10% increase in humidity, ρ decreases by 0.01, slightly enhancing the exploratory nature.

[0130] Step S604, Hardware Temperature Correction Item: ;

[0131] Physical mapping: Edge box temperature T gpu This is a hardware reliability indicator; excessively high temperatures will lead to calculation errors, particularly at the edge box temperature T. gpu When the temperature is too high, the dependence on a single solution is reduced to prevent premature convergence due to calculation errors. Therefore, the edge box temperature T... gpu When the temperature is high, the evaporation coefficient ρ should be lowered.

[0132] Derivation of hardware temperature correction term: Using 65°C as the threshold for triggering temperature adjustment, (T gpu Normalized by -65) / 10, -0.03 is the largest among all values, reflecting the high importance attached to hardware reliability. For every 10°C increase in temperature, ρ decreases by 0.03, significantly enhancing the exploration to offset potential calculation errors.

[0133] Specifically, the improved ant colony algorithm refers to a metaheuristic algorithm optimized for the distribution network reconfiguration problem. The number of network nodes N refers to the total number of electrical nodes such as all buses and load connection points in the current distribution network topology. The independent variable refers to a variable in a mathematical function whose value can be changed independently. The evaporation coefficient ρ(N) refers to a parameter controlling the pheromone evaporation rate. The pheromone concentrations H0(i,j) of all switch edges refer to the initial pheromone concentration assigned to each operable switch during algorithm initialization, with its value heuristically set based on line impedance and optimization objectives. Assignment refers to the operation of setting initial values ​​for variables during the algorithm initialization phase. Parallel release refers to the computational strategy of simultaneously simulating 32 ants independently constructing paths during algorithm iteration. The ant refers to a virtual agent in the algorithm; each ant represents an independent search process responsible for constructing a complete candidate solution. The probability... This refers to the probability that ant K will choose to operate on switch (i,j). The next switch state refers to the switch that ant is prepared to operate on next based on the current solution. The candidate reconstruction scheme S K This refers to a complete and feasible power distribution network topology constructed by a single ant. The set of operable switches refers to the set of switches that can be operated in the current topology state without causing loops or islanding in the network. The candidate reconfiguration scheme S generated by each ant... K This refers to the final network structure formed after each ant makes a series of probabilistic choices. The 2-opt switch exchange operation refers to a local search technique that generates a new neighborhood solution by randomly exchanging the states of two switches, aiming to improve the original solution. The node voltage after performing the 2-opt switch exchange operation refers to the voltage values ​​of each node obtained after performing fast power flow calculation on the new topology formed after the exchange. The network loss refers to the total active power loss caused by the resistance of each branch of the distribution network. The new scheme refers to the candidate reconfiguration scheme obtained after performing the 2-opt exchange, and the original scheme refers to the candidate reconfiguration scheme before performing the 2-opt exchange. The linearized power flow refers to the power flow calculation method used to approximately evaluate the voltage and loss of the candidate scheme. The voltage limit violation penalty Mv refers to the sum of the penalty values ​​for all nodes in the candidate scheme whose voltage exceeds the allowable range. The network loss P lossThis refers to the total active power loss calculation value corresponding to the candidate reconstruction scheme. The fitness value F of each candidate scheme is a single value used to comprehensively evaluate the merits of a candidate scheme. The higher the F value, the better the scheme is in terms of balancing voltage and loss. The global optimal path refers to the scheme with the highest fitness value F among all candidate schemes that have appeared since the algorithm started iterating. The pheromone ΔH refers to the pheromone increment added to the switch edges included in the global optimal path during the pheromone update phase. The current optimal fitness value refers to the value with the highest fitness among all candidate reconstruction schemes constructed by ants within the current iteration cycle of the improved ant colony solver. The adaptive evaporation coefficient ρ(N,t) is the evaporation coefficient that is a function of both the number of nodes N and the iteration number t. The pheromone matrix is ​​a two-dimensional array storing the current pheromone concentration of all switch edges. The improvement rate R of the best solution over 5 consecutive generations refers to the degree of relative improvement in the quality of the population's optimal solution during 5 consecutive generations of evolution in the multi-objective evolutionary algorithm. The optimal switching action table S* is... This refers to the final result output when the algorithm terminates. This embodiment does not limit the method of acquiring environmental and hardware data. For example, environmental and hardware data can be acquired through environmental sensors deployed in substations and the health monitoring interface of the edge computing device itself. The real-time solar radiation ratio SR refers to the ratio of the current real-time solar radiation intensity to the rated solar radiation intensity. The wind speed WS refers to the current real-time wind speed in the environment, in meters per second. The relative humidity RH refers to the current real-time relative humidity in the environment, in percentage. The edge box temperature Tgpu refers to the core temperature of the graphics processor or central processing unit in the edge computing device executing the algorithm. The evaporation coefficient effective probability evaluation model refers to a pre-trained neural network learning model that meets the requirements for evaluating the effective probability of the evaporation coefficient. The preset effective probability of the evaporation coefficient C0 refers to a pre-set value used to measure the effectiveness of the evaporation coefficient ρ(N), for example, a preset effective probability of the evaporation coefficient C0 = 0.93. e refers to the natural constant in mathematics, also known as Euler's number, an infinite non-repeating decimal, approximately 2.71828. This refers to heuristic information, representing the prior expectation level from node i operating switch (i,j) to node j. This refers to pheromone concentration, representing the intensity of historical experience accumulated at switch (i,j) and recognized by the ant colony. This refers to the pheromone concentration on all selectable switches (i,s) starting from node i. λ refers to the heuristic information on all optional switches (i, s) starting from node i. S refers to the index variable used to traverse and refer to all optional switches belonging to the set allowed and operable from node i. λ refers to the fitness weights pre-set by the administrator for calculating the fitness value F.

[0134] Specifically, step S6 involves making in-depth improvements to the traditional ant colony algorithm to efficiently solve the bi-objective optimization model of step S5. At the same time, by incorporating environmental and hardware data feedback, it enhances the adaptability and robustness in real industrial environments.

[0135] Specifically, when performing a safety check on the optimal switch action table in step S7, a power flow calculation safety check is performed on the optimal switch action table S* to obtain a safety check result. The optimal switch action table S* with a safety check result that meets the standard is encapsulated into a GOOSE control frame, and the GOOSE control frame is transmitted to the corresponding smart circuit breaker. The corresponding smart circuit breaker then performs self-healing control on the distributed smart distribution network according to the GOOSE control frame.

[0136] Specifically, the power flow calculation safety verification refers to a precise AC power flow calculation simulation performed before the optimal switch action table S* is officially issued and executed. The safety verification result being satisfactory means that the power flow calculation safety verification result shows that after executing the optimal switch action table S*, all preset safety operation constraints are met, and there is no risk of voltage exceeding limits or equipment overload. The GOOSE control frame refers to a specific data frame conforming to the IEC 61850 standard used to achieve fast message transmission within the substation. The corresponding smart circuit breaker refers to a circuit breaker that receives and executes the switch action command in the GOOSE control frame and has remote communication and electric operation functions. This embodiment does not limit the transmission method of transmitting the GOOSE control frame to the corresponding smart circuit breaker. For example, the GOOSE control frame can be transmitted to the corresponding smart circuit breaker through a 5G network slicing channel or the fiber optic Ethernet inside the substation.

[0137] Specifically, step S7 utilizes the millisecond-level transmission characteristics of GOOSE frames to achieve a secure and efficient closed loop from the computing space to the physical world.

[0138] Specifically, in step S8, when providing feedback on the self-healing control status of the distributed smart distribution network, the distributed smart distribution network operation data after the corresponding smart circuit breaker executes the operation according to the GOOSE control frame is collected in real time through the micro synchronous phasor measurement unit and the distribution terminal unit. This operation data, along with the fault signal, is input into the self-healing performance index evaluation model to obtain an evaluation score Z. The evaluation score Z is then compared with a preset evaluation score Z0. Based on the comparison result, the self-healing performance is judged, and the ant colony algorithm is optimized based on the judgment result.

[0139] When Z≥Z0, the self-healing performance is deemed satisfactory, and the ant colony solution method is not optimized.

[0140] When Z < Z0, the self-healing performance is deemed substandard. The ant colony algorithm is then optimized. The optimization involves obtaining the number of network nodes N in the ant colony algorithm and comparing it with the preset number of network nodes N0. Based on the comparison result, the network node size is determined, and the corrected evaporation coefficient ρ(N)` is adjusted accordingly.

[0141] When N < N0, the network node size is determined to be small, and the corrected evaporation coefficient ρ(N)` is not adjusted.

[0142] When N≥N0, the network node size is determined to be large-scale. The corrected evaporation coefficient ρ(N)` is adjusted, and the fault signal is input into the fault risk probability assessment model to obtain the fault risk probability score P. fault According to the failure risk probability score P fault The corrected evaporation coefficient ρ(N)` is adjusted to obtain the adjusted evaporation coefficient ρ(N)``, and the ant colony algorithm is performed based on the adjusted evaporation coefficient ρ(N)``, where:

[0143] ;

[0144] It is a step function;

[0145] In P fault When ≥0.8, ;

[0146] In P fault When <0.8, .

[0147] Specifically, the derivation process of the adjusted evaporation coefficient ρ(N)`` in step S8 is as follows:

[0148] Fault risk correction: ;

[0149] Physical mapping: Fault risk probability P fault Characterizes the severity and complexity of the current fault; high-risk faults, i.e., P... fault When the value is ≥0.8, the system is required to find the most robust recovery strategy and conduct a broader search. Therefore, the evaporation coefficient ρ should be reduced.

[0150] Derivation of the fault risk correction term: using a step function It enables on / off adjustment, taking effect immediately when the probability of failure exceeds 0.8. The coefficient -0.02 provides a fixed correction amount, explicitly reducing ρ in high-risk scenarios and enhancing exploration.

[0151] Specifically, the distributed smart distribution network operation data after the corresponding smart circuit breaker executes the operation according to the GOOSE control frame refers to the distributed smart distribution network operation data collected again by the micro synchronous phasor measurement unit and distribution terminal unit after the self-healing control action is completed and the power grid enters a new stable operating state. The self-healing performance index evaluation model refers to a preset machine learning model used to quantify the comprehensive performance of this self-healing process. The preset evaluation score Z0 refers to a pre-set performance qualification line, for example, the preset evaluation score Z0 = 0.82. The preset number of network nodes N0 refers to a threshold used to define the network scale level, for example, the preset number of network nodes N0 = 39. The fault risk probability evaluation model refers to the model that can evaluate the severity and potential impact of the fault based on the characteristics of the fault signal and output a fault risk probability score P. fault The neural network model.

[0152] Specifically, step S8 evaluates the actual effect of each self-healing action and feeds it back to the evaporation coefficient in step S6, thereby adaptively adjusting the strategy for networks of different sizes and faults of different severity, so as to continuously improve the decision-making quality and robustness of future self-healing actions.

[0153] Please see Figure 2 As shown, this is a schematic diagram of the structure of the self-healing control system for the distributed smart distribution network in this embodiment. The system includes:

[0154] The data acquisition module is used to collect real-time operation data of the distributed smart distribution network to obtain the operation data of the distributed smart distribution network.

[0155] The fault detection module is used to detect faults in the operation data of the distributed smart distribution network and obtain fault signals. The fault detection module is connected to the data acquisition module.

[0156] The parallel control module is used to perform parallel protection control on the distributed smart distribution network based on fault signals. The parallel control module is connected to the fault detection module.

[0157] The segment islanding module is used to construct the fault segment location model and the island partitioning model. It performs fault segment location based on the fault segment location model and island partitioning based on the island partitioning model to obtain the fault edge dataset and the island cluster dataset. The segment islanding module is connected to the parallel control module.

[0158] The ant colony optimization module is used to construct a dual-objective optimization model, solve the dual-objective optimization model according to the improved ant colony algorithm to obtain the optimal switching action table, perform safety verification on the optimal switching action table, and perform self-healing control on the distributed smart distribution network based on the safety verification results. The ant colony optimization module is connected to the section island module.

[0159] The self-healing feedback module is used to provide feedback on the self-healing control status of the distributed smart distribution network. The self-healing feedback module is connected to the ant colony optimization module.

[0160] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

Claims

1. A self-healing control method for a distributed smart distribution network, characterized in that, The method includes: Step S1: Real-time collection of distributed smart distribution network operation data to obtain distributed smart distribution network operation data; Step S2: Perform fault detection on the distributed smart distribution network operation data to obtain fault signals; Step S3: Perform parallel protection control on the distributed smart distribution network based on the fault signal; Step S4: Construct a fault segment localization model and an island partitioning model, locate fault segments according to the fault segment localization model, partition islands according to the island partitioning model, and obtain a fault edge dataset and an island cluster dataset. Step S5: Construct a dual-objective optimization model and use the fault edge dataset and the islandable cluster dataset as inputs to the dual-objective optimization model to obtain the dual-objective optimization model. The dual-objective optimization model takes the minimum node voltage deviation and the minimum network loss after the fault as the optimization objectives, and the branch current does not exceed the thermal stability limit, the node voltage is maintained between 0.95pu and 1.05pu, the output of the distributed power source does not exceed its rated capacity, and the number of switching actions is limited as constraints. Step S6: Solve the bi-objective optimization model using the improved ant colony algorithm to obtain the optimal switching action table. The improved ant colony algorithm includes: calculating the evaporation coefficient with the number of network nodes as the independent variable; releasing multiple ants in parallel to select the next switching state according to probability to form candidate reconstruction schemes; performing 2-opt switching operations on the candidate reconstruction schemes; using linearized power flow to calculate voltage over-limit penalties and network losses and calculating fitness values ​​according to the fitness function; releasing additional pheromones for the globally optimal path and updating the pheromone matrix according to the adaptive evaporation coefficient; and outputting the optimal switching action table when the improvement rate of the optimal solution for multiple consecutive generations is lower than a threshold. The improved ant colony algorithm also acquires environmental and hardware data, inputs the environmental and hardware data into the evaporation coefficient effective probability evaluation model to obtain the effective probability of the evaporation coefficient, corrects the evaporation coefficient based on the effective probability, and performs ant colony solving based on the corrected evaporation coefficient. Step S7: Perform a safety check on the optimal switch action table, and perform self-healing control on the distributed smart distribution network based on the safety check results; Step S8: Feedback is provided on the self-healing control status of the distributed smart distribution network. The distributed smart distribution network operation data after the operation is executed is collected in real time through the micro synchronous phasor measurement unit and the distribution terminal unit. The distributed smart distribution network operation data after the operation is executed and the fault signal are input into the self-healing performance index evaluation model to obtain the evaluation score. The ant colony solution method is optimized based on the evaluation score. The optimization method is to adjust the corrected evaporation coefficient according to the number of network nodes and the fault risk probability score, and then execute the ant colony solution method according to the adjusted evaporation coefficient. In step S3, when performing parallel protection control of the distributed smart distribution network based on the fault signal, the fault signal is input into the fault current identification model for fault current value identification to obtain the fault current value Ig. The fault current value Ig is compared with the preset fault current value Ig0. Based on the comparison result, the emergency situation of the fault is judged, and the parallel protection control strategy is output based on the judgment result, wherein: When Ig≥Ig0, the fault emergency situation is determined to be emergency, and the output parallel protection control strategy is to issue a trip command; When Ig < Ig0, the emergency situation is determined to be non-emergency, and the output parallel protection control strategy is to issue a deep reinforcement learning control command.

2. The self-healing control method for distributed smart distribution networks according to claim 1, characterized in that, In step S2, when performing fault detection on the distributed smart distribution network operation data, the distributed smart distribution network operation data is input into the fault detection model. The fault detection model outputs current surge I2 and voltage sag value U2, and compares the current surge I2 and voltage sag value U2 with preset current surge I0 and preset voltage sag value U0, respectively. Based on the comparison results, the fault occurrence is judged, and a fault signal is output based on the judgment result, wherein: When the comparison result is I2≤I0 and U2≥U0, the fault occurrence is determined to be no fault, the subsequent operation is stopped, and steps S1-S2 are repeated until the fault occurrence is a fault, at which point the distributed smart distribution network operation data corresponding to the fault occurrence is output as a fault signal. When the comparison result is not I2≤I0 and U2≥U0, the fault situation is determined to be a fault, and the distributed smart distribution network operation data corresponding to the fault situation is output as a fault signal.

3. The self-healing control method for distributed smart distribution networks according to claim 1, characterized in that, In step S3, the deep reinforcement learning control instructions include: inputting the fault signal into the fault instantaneous trip model, outputting the instantaneous trip setting value of the line protection from the fault instantaneous trip model, and transmitting the instantaneous trip setting value of the line protection to the digital protection relay.

4. The self-healing control method for distributed smart distribution networks according to claim 1, characterized in that, In step S4, a fault segment location model is constructed. When locating fault segments based on the fault segment location model, the historical fault segment location database is divided into an 80% segment training set and a 20% segment test set. The segment training set is input into a graph neural network model to train the graph neural network model, resulting in a trained graph neural network model. The segment test set is input into the trained graph neural network model for testing until the segment testing accuracy Q of the trained graph neural network model reaches 95%. Then, the trained graph neural network model is output as the fault segment location model, and the fault signal is input into the fault segment location model to obtain the fault edge dataset.

5. The self-healing control method for distributed smart distribution networks according to claim 4, characterized in that, In step S4, an island partitioning model is constructed. When partitioning islands according to the island partitioning model, the island partitioning database is divided into a 70% island training set, a 15% island validation set, and a 15% island test set. The island training set is input into a convolutional neural network model to train the convolutional neural network model, resulting in a trained convolutional neural network model. The island validation set is input into the trained convolutional neural network model for iterative optimization, resulting in a validated convolutional neural network model. The island test set is then input into the validated convolutional neural network model until the island test accuracy G of the validated convolutional neural network model reaches 98%. At this point, the validated convolutional neural network model is output as the fault island partitioning model, and the distributed smart distribution network operation data is input into the fault island partitioning model to obtain a dataset of possible island clusters.

6. The self-healing control method for distributed smart distribution networks according to claim 1, characterized in that, In step S5, when constructing the dual-objective optimization model, "minimum node voltage deviation after fault" and "minimum network loss" are taken as the optimization objectives of the dual-objective optimization model. "Branch current does not exceed the thermal stability limit", "node voltage is maintained between 0.95pu and 1.05pu", "distributed power output does not exceed its rated capacity", and "the number of switching actions k is limited to k≤K0 times" are taken as the constraints of the dual-objective optimization model. The fault edge dataset and the islandable cluster dataset are taken as the input of the dual-objective optimization model to obtain the dual-objective optimization model. The dual-objective optimization model is the mathematical optimization model to be solved, and K0 is the preset limit number of times.

7. The self-healing control method for distributed smart distribution networks according to claim 1, characterized in that, When solving the bi-objective optimization model using the improved ant colony algorithm in step S6, the bi-objective optimization model is solved using the ant colony solution method, which includes: Step Y01, Initialization Phase: Using the number of network nodes N as the independent variable, according to ρ(N) = 0.90 – 0.40·e (–N / 50) Calculate the evaporation coefficient ρ(N) and assign values ​​to the pheromone H0(i,j) of all switch edges; Step Y02, Ant Path Construction Phase: Release 32 ants in parallel, according to probability. Select the next switch state to form a candidate reconstruction scheme S K ,in, , allowed is the set of switches that are allowed to operate; Step Y03, Local Search Phase: Generating candidate reconstruction schemes S for each ant K If the node voltage increases and the network loss decreases after performing the 2-opt switch operation, then the new solution replaces the original solution. Step Y04, Fitness Evaluation Phase: Linearized power flow calculation is used to calculate the voltage over-limit penalty Mv and network loss P. loss And according to the fitness function F=λ·Mv+(1-λ)·P loss Calculate the fitness value F for each candidate solution, where λ refers to the fitness weight pre-set by the administrator for calculating the fitness value F, and λ∈[0,1]. Step Y05, Pheromone Update Phase: Additional pheromone ΔH is released for the globally optimal path, and ΔH = Q / F. best Where Q is a constant, F best The current optimal fitness value is set, and the pheromone matrix is ​​updated according to the adaptive evaporation coefficient ρ(N,t); Step Y06, Termination Judgment Stage: When the improvement rate R of the best solution for 5 consecutive generations is less than 0.1%, output the optimal switching action table S*.

8. The self-healing control method for distributed smart distribution networks according to claim 7, characterized in that, In step S6, when solving the bi-objective optimization model using the improved ant colony algorithm, environmental data and hardware data are also acquired. The environmental data includes real-time solar radiation ratio (SR), wind speed (WS), and relative humidity (RH). The hardware data includes edge box temperature (T). gpu The environmental and hardware data are input into the evaporation coefficient effective probability evaluation model to obtain the evaporation coefficient effective probability C. The evaporation coefficient effective probability C is compared with the preset evaporation coefficient effective probability C0. Based on the comparison result, the effectiveness of the evaporation coefficient is judged, and the evaporation coefficient ρ(N) is corrected according to the judgment result. When C≥C0, the evaporation coefficient is considered valid and no correction is made to the evaporation coefficient ρ(N); When C < C0, the effective case of the evaporation coefficient is determined to be invalid, and the evaporation coefficient ρ(N) is corrected. The corrected evaporation coefficient is then set to ρ(N)'. And the ant colony method is executed based on the corrected evaporation coefficient ρ(N)`.

9. The self-healing control method for distributed smart distribution networks according to claim 1, characterized in that, When providing feedback on the self-healing control status of the distributed smart distribution network in step S8, the distributed smart distribution network operation data after the corresponding smart circuit breaker executes the operation according to the GOOSE control frame is collected in real time through the micro synchronous phasor measurement unit and the distribution terminal unit. The distributed smart distribution network operation data after the corresponding smart circuit breaker executes the operation according to the GOOSE control frame, along with the fault signal, are input into the self-healing performance index evaluation model to obtain the evaluation score Z. The evaluation score Z is compared with the preset evaluation score Z0. Based on the comparison result, the self-healing performance is judged, and the ant colony solution method is optimized based on the judgment result. Wherein: When Z≥Z0, the self-healing performance is deemed satisfactory, and the ant colony solution method is not optimized. When Z < Z0, the self-healing performance is deemed substandard. The ant colony algorithm is then optimized. The optimization involves obtaining the number of network nodes N in the ant colony algorithm and comparing it with the preset number of network nodes N0. Based on the comparison result, the network node size is determined, and the corrected evaporation coefficient ρ(N)` is adjusted accordingly. When N < N0, the network node size is determined to be small, and the corrected evaporation coefficient ρ(N)` is not adjusted. When N≥N0, the network node size is determined to be large-scale. The corrected evaporation coefficient ρ(N)` is adjusted, and the fault signal is input into the fault risk probability assessment model to obtain the fault risk probability score P. fault According to the failure risk probability score P fault The corrected evaporation coefficient ρ(N)` is adjusted to obtain the adjusted evaporation coefficient ρ(N)``, and the ant colony algorithm is performed based on the adjusted evaporation coefficient ρ(N)``, where: ； It is a step function; In P fault When ≥0.8, ; In P fault When <0.8, .

10. A system applied to the self-healing control method for distributed smart distribution networks as described in any one of claims 1-9, characterized in that, The system includes: The data acquisition module is used to collect real-time operation data of the distributed smart distribution network to obtain the distributed smart distribution network operation data; The fault detection module is used to detect faults in the operation data of the distributed smart distribution network and obtain fault signals. The parallel control module is used to perform parallel protection control on the distributed smart distribution network based on fault signals. The segment island module is used to construct a fault segment location model and an island partitioning model. It performs fault segment location based on the fault segment location model and island partitioning based on the island partitioning model, resulting in a fault edge dataset and a potentially isolated cluster dataset. The ant colony optimization module is used to construct a dual-objective optimization model, solve the dual-objective optimization model according to an improved ant colony algorithm to obtain the optimal switching action table, and perform safety verification on the optimal switching action table. Based on the safety verification results, it performs self-healing control on the distributed smart distribution network. The improved ant colony algorithm includes: calculating the evaporation coefficient with the number of network nodes as the independent variable; releasing multiple ants in parallel to select the next switching state according to probability to form candidate reconfiguration schemes; performing 2-opt switching operations on the candidate reconfiguration schemes; using linearized power flow to calculate voltage over-limit penalties and network losses and calculating fitness values ​​according to the fitness function; releasing additional pheromones for the globally optimal path and updating the pheromone matrix according to the adaptive evaporation coefficient; and outputting the optimal switching action table when the improvement rate of the optimal solution for multiple consecutive generations is lower than a threshold. The improved ant colony algorithm also acquires environmental data and hardware data, inputs the environmental data and hardware data into the evaporation coefficient effective probability evaluation model to obtain the effective probability of the evaporation coefficient, corrects the evaporation coefficient based on the effective probability, and performs ant colony solving based on the corrected evaporation coefficient. The self-healing feedback module is used to provide feedback on the self-healing control status of the distributed smart distribution network. It collects the distributed smart distribution network operation data in real time after the operation is executed through the micro synchronous phasor measurement unit and the distribution terminal unit. The distributed smart distribution network operation data after the operation is executed and the fault signal are input into the self-healing performance index evaluation model to obtain the evaluation score. The ant colony solution method is optimized based on the evaluation score. The optimization method is to adjust the corrected evaporation coefficient according to the number of network nodes and the fault risk probability score, and then execute the ant colony solution method based on the adjusted evaporation coefficient.

Images